Four Thousand Exception Messages a Week After Go-Live
Six weeks after an ERP go-live, a discrete manufacturer with 1,800 purchased part numbers is receiving about 4,200 MRP action messages per week: roughly 1,900 reschedule-in, 1,100 expedite, 700 cancel and 500 release. Planners have stopped working the list and instead react to the loudest buyer. On-time delivery is 71% and falling, and the plant is expediting freight daily. The data audit shows: every purchased item was loaded with a 30-day planned delivery time because the migration template defaulted it; actual lead times range from 14 days for bearings to 90 days for castings; safety stock is zero on all items; BOM scrap is zero everywhere even though machining yield runs about 5% scrap; and the forecast was loaded at 100% of expected demand while the open order book was also loaded, with forecast consumption not configured.
1. How should the planner triage 4,200 messages rather than working top to bottom? Give a concrete filter sequence and the reasoning behind it.
Read the worked answer
Working 4,200 messages sequentially guarantees that the planner spends the week on low-value noise while a $40,000 casting order goes late. Triage by consequence and by confidence in the message.
Sequence:
- Root-cause clusters first. Group messages by item and by message type. If 900 reschedule-ins trace to the same wrong lead time or the same mis-set lot size, one data fix removes 900 messages. Fixing the cause beats working the symptom 900 times.
- Dollar value and constraint items. Act on messages for A-class parts, long-lead castings, and anything feeding a constraint work centre or a customer order inside the frozen window. Defer C-class messages that carry no near-term requirement.
- Date proximity inside the current lead time. An expedite for a part needed in 5 days with a 14-day lead time is a real decision now. An expedite for something needed in 11 weeks is a planning artifact - validate the date before calling the supplier.
- Trust check per message type. Reschedule-in and expedite assume the dates in the system, which the lead time audit has already shown are wrong. Until lead times are corrected, treat exception dates as suspect; treat cancel messages with extreme care, because cancelling real supply on bad data is how shortages are created.
Target: roughly 300-400 messages actioned per week after the first data fix, with a documented rule for what is deferred and why. Publish the count of messages actioned within three days as a metric, not the raw count of messages received.
2. Quantify the effect of the lead-time and forecast errors, and explain why the message mix is dominated by reschedule-in and expedite rather than cancel.
Read the worked answer
Lead time. A casting with a true 90-day lead time loaded as 30 days is planned to be released 60 days late, every time, for every part number involved. If 40 casting part numbers each need three orders a year, that is about 120 orders a year that are born late - and every one produces expedite messages, premium freight and a supplier asked to compress a 90-day process into 30. Conversely, bearings with a true 14-day lead time loaded as 30 days are released 16 days too early: too much inventory, and reschedule-in messages when the need date moves.
Forecast plus orders. Loading 100% of forecast and also the open order book, with no consumption, roughly doubles demand wherever the two overlap. If 70% of the forecast period is already covered by booked orders, the plan carries up to 170% of real demand - inflated gross requirements, inflated lot sizes, and a wave of cancel and reschedule messages in two to three weeks when reality arrives.
Why reschedule-in and expedite dominate: with a 30-day master for 90-day parts, the system believes the material can be obtained later than it can. So need dates are always 'sooner than the plan thought' - reschedule-in. The 700 cancels are the residue of the inflated forecast now being corrected. The mix is a fingerprint of understated lead times plus overstated demand, not of unstable customer demand.
3. Which metrics and data-governance routines should be put in place so that message volume becomes a managed signal rather than a crisis?
Read the worked answer
Data accuracy, measured on samples and owned by name:
- Item master lead time accuracy (sample of 50-100 parts per month vs actual PO receipt history; target 95% plus, with a named owner in purchasing).
- Inventory record accuracy by location and cycle count (target 95-98% by count and by value).
- BOM accuracy, including scrap and effectivity (audit the top 20% of parts by value or usage).
- Forecast consumption configuration verified once, then tested whenever forecast or order interfaces change.
Process stability:
- Exception messages per planner per week, and percent actioned within three days. Rising volume with flat demand means data or parameters, not workload.
- MPS adherence inside the frozen window (target 95% plus); every unauthorized change logged with a reason code.
- Schedule attainment and on-time delivery, split by cause - material, capacity, quality - so the ERP discussion stays factual.
- Number of expedites and premium-freight dollars per month as the financial proxy for planning quality.
Governance: a weekly 30-minute data review with purchasing, planning, engineering and IT; a change-control rule for item master and BOM fields with an audit trail; mass-update scripts reviewed before execution; and a stated target for message volume (for example, under 20% of purchased items generating a message in a week) so that the team is managing the plan rather than the inbox.